Nasiko-Labs/nasiko

What it solves

Nasiko provides a centralized control plane for managing fleets of AI agents. It addresses the operational complexities of agent-to-agent (A2A) communication, such as routing traffic, managing API keys, securing agent access, and monitoring costs and performance without requiring agents to be publicly exposed.

How it works

Nasiko acts as a single process that sits in front of all agents. It terminates TLS and proxies all inter-agent traffic, ensuring that agents are never publicly reachable. It uses a three-stage routing engine (embedding similarity, reranking, and LLM selection) to direct requests to the correct agent. To enhance security, it provides an LLM Router that gives agents short-lived identity tokens instead of real API keys, and an MCP Gateway that manages tool credentials centrally.

Who it’s for

Developers and operators building multi-agent systems who need a secure, observable, and scalable way to deploy and route traffic between agents written in different languages (Python, Rust, Go, TypeScript) that follow the A2A protocol.

Highlights

• A2A Protocol Support: Deploy any agent that speaks the A2A v1.0 spec regardless of language.
• Intelligent Routing: A 3-stage pipeline for agent selection based on context and embeddings.
• Security-First Architecture: Agents are proxied and never public; secrets are AES-256-GCM encrypted at rest.
• Built-in Observability: Full OTel integration for end-to-end tracing, token usage, and cost tracking.
• Embedded OCI Registry: Self-hosted, S3-backed registry for building and deploying agent images without external dependencies.
• Flow Guards: Redis-backed limits on depth, fan-out, and token budgets to prevent runaway agent loops.